)]}'
{
  "commit": "3a18feca624cd35bd84d4ae594e5a095898a840e",
  "tree": "8700c7b0a3a71d91629c7857bf36b6faaed49064",
  "parents": [
    "a510f030095462bf0ca623018e580adeb110279d"
  ],
  "author": {
    "name": "Keith Lee",
    "email": "leekei@apache.org",
    "time": "Fri Jun 12 23:40:52 2026 +0100"
  },
  "committer": {
    "name": "Roman",
    "email": "khachatryan.roman@gmail.com",
    "time": "Wed Jul 08 00:01:34 2026 +0200"
  },
  "message": "[FLINK-39924] Size jemalloc narenas from container CPU allowance\n\nApache Flink containers load jemalloc via LD_PRELOAD but don\u0027t configure\nnarenas. jemalloc\u0027s default is 4 * ncpus, where ncpus is read from\n/proc/cpuinfo, the host CPU count, not the container\u0027s CPU limit. In\nCPU limited pods on large hosts this over-provisions arenas and causes\nRSS fragmentation, since each arena holds dirty pages for dirty_decay_ms\nbefore releasing them to the OS.\n\nDetermine the effective CPU count from the cgroup CPU quota directly\n(cpu.max for v2, cpu.cfs_quota_us / cpu.cfs_period_us for v1), since\nnproc honors cpuset but not CPU quotas. Fall back to nproc when no quota\nis set. Skip the override entirely when the user has supplied narenas\nin MALLOC_CONF, and append narenas to any other user-supplied MALLOC_CONF\nvalue.\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "cf63daa2c591f9d64e10c11c894f2b9b674a6b4d",
      "old_mode": 33261,
      "old_path": "docker-entrypoint.sh",
      "new_id": "8accfbc8e389f1b36d10353bcaf87ed9f6435a1d",
      "new_mode": 33261,
      "new_path": "docker-entrypoint.sh"
    }
  ]
}